Extreme's Markus Nispel On Agent Governance: 3 Controls For Production Autonomy

Описание: Extreme Networks deployed autonomous AI agents in production network infrastructure using a three-layer governance model that solves the trust problem: deployment scope, action scope, and autonomy controls. Markus Nispel, CTO EMEA and Head of AI Engineering, breaks down their evolution from 2018 AI ops to multi-agent systems that analyze event correlations impossible for humans and auto-generate support tickets.

Their ARC framework (Acceleration, Replacement, Creation) separates mandatory automation from competitive differentiation, focusing investment on creation use cases where ROI discussions become simpler. They reduced time-to-knowledge by 90% through RAG systems that let new engineers access tens of thousands of documentation pages instantly, addressing critical talent shortages in network operations.

CHAPTERS:
00:00 Introduction
01:13 Extreme Networks background and AI evolution from 2018
02:56 Early AI and machine learning use cases in network operations
04:55 Challenges bringing generative AI to deterministic infrastructure
08:17 The ARC framework: Acceleration, Replacement, and Creation
09:12 Managing complexity and POC-to-production challenges
11:36 Prioritizing use cases and getting stakeholder buy-in
12:57 Creation use cases: Network operations and troubleshooting
16:06 90% time-to-knowledge reduction through RAG systems
17:27 Building trust through accuracy and transparency
18:01 Agent governance: Explainability, transparency, and control
19:57 Three-layer governance model: Deployment, action, autonomy
21:40 Managing pilot-to-production deployment at scale
23:24 Measuring accuracy: Pre-production vs production metrics
24:37 Evaluation challenges and emerging business opportunities
26:30 Pre-production SME validation and GTAC collaboration
27:28 AI team structure and organizational evolution
29:32 The AI mesh: Distributed ownership with central architecture
31:24 Data diversity and cross-domain agent collaboration
31:46 Build vs buy decisions for AI systems
33:41 Priorities and challenges: Continuous rebuilding in AI
34:39 Agent interoperability: MCP, A2A, and the agentic web
36:02 AGI discussion and AI as a tool
38:29 Personal AI usage and tools
39:25 Resources for staying current in AI
40:16 Key concepts: AI replacing humans and abundance
42:24 Closing and where to follow Markus